Finite difference and finite volume methods for transport and

conservation laws

Boualem Khouider

PIMS summer school on stochastic and probabilistic methods for atmosphere, ocean, and
dynamics.
University of Victoria, July 14-18, 2008.

Contents

1 Introduction to finite differences: The heat equation 4

1.1 Explicit scheme for the heat equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2 Stability of the forward scheme: von Neumann analysis . . . . . . . . . . . . . . . . 9

1.3 Implicit scheme for the heat equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.4 The Crank-Nicholson scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Time splitting methods 12

3 Introduction to quasi-linear equations and scalar conservation laws 14

3.1 Prototype examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.2 Solutions by the method of characteristics . . . . . . . . . . . . . . . . . . . . . . . . 17

3.3 Notion of shocks and weak solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3.4 Discontinuous initial data and the Riemann problem . . . . . . . . . . . . . . . . . . 21

3.5 Non-uniqueness of weak solutions and the entropy condition . . . . . . . . . . . . . . 24

1
4 Finite difference schemes for the advection equation 26

4.1 Some simple basic schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.2 Accuracy and consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4.3 Stability and convergence: the CFL condition and Lax-equivalence theorem . . . . . 28

4.4 More on the leap-frog scheme: the parasitic mode and the Robert-Asselin filter . . . 32

4.5 The Lax-Friedrichs scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.6 Second order schemes: the Lax-Wendroff scheme . . . . . . . . . . . . . . . . . . . . 34

4.7 Some numerical experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.8 Numerical diffusion, dispersion, and the modified equation . . . . . . . . . . . . . . . 38

5 Finite volume methods for scalar conservation laws 43

5.1 Wrong shock speed and importance of conservative form . . . . . . . . . . . . . . . . 43

5.2 Godonuvs first order scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

5.3 High resolution, TVD, and MUSCL schemes . . . . . . . . . . . . . . . . . . . . . . . 47

2
Foreword

The celebrated Chapman-Kolmogorov equation for a diffusion Markovian process reduces to the
well known Fokker-Planck equation [Gardiner, 2004]

p(z, t/y, t ) 1 X 2
+ z (A(z, t)p(z, t/y, t)) = (Bi,j (z, t)p(z, t/y, t)) . (1)
t 2 zi zj
i,j

Here p(z, t/y, t ) is the probability density distribution of the random variable z at time t given y at
time t of the underlying Markovian process. z is the gradient differential operator with respect
to the variable z, A(z, t) is a vector function known as the drift, representing the deterministic
dynamics of the process, and B = [Bi,j ] is the diffusion matrix describing the Gaussianity or
randomness of the process. When B = 0 the Focker-Planck equation is also known as the Liouville
equation, describing the evolution of the probability distribution of a random process undertaking
deterministic dynamics. Taking the derivative with respect to the known variable y at time t ,
instead, yields the famous backward equation [Gardiner 2004]

p(x, t/y, t ) 1X 2
p(x, t/y, t ) .



+ A(y, t ) y p(x, t/y, t ) = B i,j (y, t ) (2)
t 2 yi yj
i,j

Note that while the two partial differential equations above are given in terms of the variables
z and t or y and t , respectively, the remaining variables (y, t ) for the first equation and (x, t)
for the second) can be treated as parameters and therefore ignored when we are only concerned
with numerical or analytic solution methodology for theses PDEs. This kind of equations are
wide spread in the applied physical sciences. The term involving the vector A is also known as a
transport process. In fluid mechanics, for example, it models the action of the flow field on the
dynamical quantity under consideration, such as temperature, density, or momentum. It appears
in either a conservative form
t q + (Aq) = 0 (3)
as in the forward Focker-Plank equation (1) or in advective form

t q + A q = 0 (4)

as in the backward equation (2). Also under the obvious condition of ellipticity, the matrix B
can be diagonalized and the associated diffusion operator is reduced to the more standard Laplace
operator. Therefore, we are interested here in the numerical solution of the advection diffusion
equation
t q + A q = Dq (5)
or
t q + (Aq) = Dq (6)
which is a superposition of the transport equation of conservative or advective type and the diffusion
equation
ut = Du
also known as the heat equation. In this series of lectures we will discuss some standard numerical
technics for these types of equations. Special emphasis will be given to finite difference and finite

3
volume methods for the advection and conservation equations in (4) and (3), respectively. We will
treat in some details the case when the advection field A depends on the solution q that leads to
shock formation and other types of singularities, which are important in gas dynamics and other
fields of practical importance.

Unless otherwise stated, from now on, we consider only partial differential equations in the 2
variables (x, t), where < x < + represents the space variable and t 0 is time. The solution
is denoted by u(x, t).

1 Introduction to finite differences: The heat equation

We introduce some basics of the finite difference methodology for partial differential equation
through the simple case of the heat or diffusion equation in 1 dimension
ut = Duxx ,
where D > 0 is a constant heat conduction or diffusion coefficient.

The finite differences method applied to the heat equation above, starts by the approximation of
the partial derivatives, ut and uxx by their corresponding finite difference quotients. For a smooth
function f (x) of the variable x, we have according to Taylor expansion
1 1 1
f (x0 + h) = f (x0 ) + f (x0 )h + f (x0 )h2 + + f (n) (x0 )hn + f (n+1) ()h( n + 1)
2 n! (n + 1)!
where h is a non zero increment or displacement along the real line, starting from a fixed point
x0 and is between x0 and x0 + h. Recall that, a function g is said a big O of h and we write
g = O(hp ) if
g(h)
lim = Constant .
h0 hp

Assuming h > 0 and using the Taylor approximation for f (x0 + h) and f (x0 h), the forward and
backward difference formulas follow immediately,
Forward Formula:
f (x0 + h) f (x0 ) 1 f (x0 + h) f (x0 ) f (x0 + h) f (x0 )
f (x0 ) = f ()h = + O(h) , (7)
h 2 h h
Backward Formula:
f (x0 ) f (x0 h) 1 f (x0 ) f (x0 h) f (x0 ) f (x0 h)
f (x0 ) = + f ()h = + O(h) , (8)
h 2 h h

Furthermore, the 3rd order Taylor approximations of the difference f (x0 + h) f (x0 h) yields the
Centered Formula:
f (x0 + h) f (x0 h) 1 f (1 ) + f (2 ) 2
f (x0 ) = h
2h 6 2
f (x0 + h) f (x0 h) f (x0 + h) f (x0 h)
= + O(h2 ) , (9)
2h 2h

4
where x0 h 1 x0 2 x0 + h, whereas the 4th order Taylor approximation of the sum
f (x0 + h) + f (x0 h) leads to an approximation for the second order derivative f (x0 ).

Centered Formula for the second order derivative:

f (x0 + h) 2f (x0 ) + f (x0 h) 1 (
f (x0 ) = + ( (1 ) + f (2 ))h2
h2 24 f
f (x0 + h) 2f (x0 ) + f (x0 h) f (x0 + h) 2f (x0 ) + f (x0 h)
= 2
+ O(h2 ) , (10)
h h2
The formulas on the very-right hand sides of (7) to (10) are only some of the very basic examples of
finite difference approximations for the first and second order derivatives to a first and second order
accuracy, respectively. ( The forward and backward finite difference approximations in (7) and (8)
are first order accurate, therefore called first order approximations while those in (9) and (10) are
second order accurate and are called second order approximations. ) Finite difference formulas of
higher order and for higher order derivative can be derived by using similar manipulations of Taylor
approximations or polynomial approximations (e.g. interpolation). Also different combinations of
points to the left or the right of the point x0 can be considered separately.

A finite difference method for a given partial differential equation PDE consists of the approximation
of the partial derivatives of its (unknown) solution u by a corresponding finite difference formula
of a certain order.

1.1 Explicit scheme for the heat equation

Consider the heat equation

ut = Duxx
on a finite rod x (0, L) with the initial condition u(x, 0) = u0 (x) and boundary conditions
u(0, t) = (t) and u(L, t) = (t), t [0, T ]. Consider a discretization of the rectangle [0, L] [0, T ]
into a finite number of nodes (xj , tn ), j = 0, 1, , M + 1, n = 0, 1, , N such that xj = jh and
tn = nt where h x = L/(M + 1) and t = T /N . The set of all points (xj , tn ) is called
a grid or a mesh while h and t are respectively called the time step and spatial grid size. Let
unj = u(xj , tn ). Using a forward finite difference approximation for the time derivative combined
with a centered formula for the second order spacial derivative, applied at each node (j, n), the
heat equation can be rewritten as

un+1
j unj unj+1 2unj + unj1
+ O(t) = D + O(h2 ). (11)
t h2
Let wjn be the finite sequence of real numbers satisfying the following difference scheme

Dt n
wjn+1 = wjn + (wj+1 2wjn + wj1
n
), j = 1, , M, n = 0, 1, N 1, (12)
h2
obtained from (11) by dropping the small error terms O(t) and O(h2 ), known as the truncation
error, and using the initial and boundary conditions

wj0 = u0 (xj ), w0n = (tn ), wM

n
+1 = (tn ).

5
This is the main philosophy behind finite differences of obtaining an approximate solution for the
given PDE at the interior grid points

u(xj , tn ) wjn , j = 1, , M, n = 1, 2 , N.

As we shall see below, for the scheme (12) for the heat equation, we have

u(xj , tn ) = wjn + O(t) + O(h2 ),

i.e, the approximation is first order in time and second order in space, which is a statement of
convergence as well, as t, h 0, which is proved below after the introduction of the notions
of consistency and accuracy. For a given finite discretization, an approximation for the solution
u(x, t) at a given interior point (x, t) (0, L) (0, T )not on part of the grid can be obtained by 2d
interpolation of the discrete solution wjn . Ideally, the order of interpolation would match that of the
numerical approximation to obtain an optimal approximation in terms of efficiency and accuracy.

Note that the initial condition u0 (x) and the boundary conditions (t), (t) provide the starting
n , n = 1, 2, , N
points wj0 , j = 0, 1, , M + 1 and the lateral grid point values w0n and wM
respectively while the numerical scheme is evolved from time step to time step using the formula
(12) to provide the solution at time tn + t, which is given explicitly in terms of the solution at
time tn . Such a scheme is called explicit as opposed to an implicit scheme where obtaining wjn+1
from wjn involves the inversion of a linear or non-linear system of algebraic equations (see next
subsection).

Definition 1 (Consistency):
A numerical scheme, L(wjn ) = 0, for a given PDE, P(u(x, t)) = 0, is said consistent if the trunca-
tion error
h (h, t) P(u(x, t)) L(unj ) 0, h, t 0.
The scheme is said consistent of order (p, q) if

h (h, t) = O(hp ) + O(tq ).

For the forward explicit scheme (12) for the heat equation we have

un+1
!
j unj unj+1 2unj + unj1
ut Duxx D = O(h2 ) + O(t),
t h2

i.e, the forward scheme for the heat equation is consistent of order (2, 1).

Definition 2 (Stability):
A numerical scheme, L(wjn ) = 0 for an evolution partial differential equation on [0, T ] is said stable
if the discrete solution satisfies

max |wjn | C, n = 1, , N
j=1, ,M

where C > 0 is a constant independent on the grid size, h, t.

6
Theorem 1 (Convergence, Lax equivalence theorem)
If the original PDE problem is well posed then, the discrete solution of the numerical scheme
converges to the solution of the PDE when h, t 0 if and only if the scheme is consistent and
stable.

Proof:
For simplicity in exposition, we assume that both the PDE and the numerical scheme are linear.
Let u(x, t) be the the solutions to the corresponding PDE and wjn the discrete solution of the
numerical scheme. Let unj = u(xj , tn ). The linear numerical scheme can be written as

wn+1 = Lh,t wn

where wn = (wjn ) is the Rn vector representing the discrete solution and Lh,t is the linear operator
(a matrix) associated with the numerical scheme, which depends on the grid size parameters h and
t. For the forward scheme (12) for the heat equation, we have
Dt n
(Lh,t wn )j = wjn + (wj+1 2wjn wj1
n
).
h2

Note that the stability requirement implies that the operator L and all its powers stay bounded as
N, M + (or equivalently as h, t 0), i.e,

Stability = ||Lh,t ||n C, N, M, n 0, n N.

To satisfy such condition if suffices to have

||Lh,t || 1.

However this later condition can be relaxed to

||Lh,t || 1 + O(t).

By the consistency requirement, we have

un+1 = Lh,tun + O(t2 ) + O(h2 )t.

Thus,
un+1 wn+1 = Lh,t(un wn ) + O(t2 ) + O(h2 )t.
Using basic linear algebra we have

||un+1 wn+1 || ||Lh,t ||||un wn || + O(t2 ) + O(h2 )t,

and by induction on n, we arrive to

||un+1 wn+1 || ||Lh,t ||n+1 ||u0 w0 ||+(||Lh,t ||n +||Lh,t ||n1 + +||Lh,t||+1)(O(t2 )+O(h2 )t).

From the initial condition, we have u0 w0 = 0. Thus,

||un+1 wn+1 || ((1 + O(t))n + (1 + O(t))n1 + + (1 + O(t)) + 1)(O(t2 ) + O(h2 )t)

7
 (1 + O(t))n+1 1
(O(t2 ) + O(h2 )t) (eT 1)(O(t) + O(h2 )) 0, t, h 0,
t
and the rate of convergence is the same as the order of consistency, i.e, linear in time and quadratic
in space.

Numerical Tests and stability of the forward scheme

Consider the PDE
1
ut = uxx , x (0, 1), t (0, T )
16
u( x, 0) = sin(2x) (13)
u(0, t) = u(1, t) = 0.

The exact analytical solution for this PDE is given by

1 2
u(x, t) = e 4 t sin(2x).

The matlab code for solving this problem using the forward/explicit scheme (12) is given below
in (14) and the results obtained with two different time-step sizes are plotted in Figure 13. The
spacial discretization consists of 11 grid points and the time step is t = 0.02 for the top panel and
?t = 0.2 for the bottom panel. With t = 0.02, the numerical scheme (12) provides an accurate
solution for this problem while the larger time step value t = 0.2 leads to a numerical solution
that grows without bounds. Such behavior is known as a numerical instability. In fact, we show
below that the forward scheme (12) is conditionally stable ; It is stable only for a relatively small
t values; the largest eigenvalue of the linear operator associated with the forward scheme (12) is
smaller or equal to one up to O(t) provided Dt/h2 12 . Instead of going through the tedious
task of computing the eigenvalues of the matrix, we use an alternate methodology for stability of
difference schemes known as the von Neumann analysis.

%%%Forward scheme for the heat equation:

%%% INPUT
%%%%%Advection velocity:
mu = 1/16;
%%%%Grid size; Use periodic boundary conditions
X=1;M=10;Tend=4;
h=1/(M+1);
Dt=0.02;
x= 0:h:X; %%% x(1) =0, x(2) = h, , ..., x(M) = X -h; x(M+1) = X;
wn=sin(2*pi*x);
time=0;
mu = mu*Dt/h^2;
while(time<Tend)
wn(2:M+1) = wn(2:M+1) +mu*(wn(3:M+2)-2*wn(2:M+1)+wn(1:M));
time=time+Dt;
end
figure(2)
xx=0:1/1000:1;
plot(xx, exp(-time/4*pi^2)*sin(2*pi*xx))

8
 hold on
plot(x,wn,x,linewidth,2)

Matlab code for the explicit scheme for the heat equation. (14)

1.2 Stability of the forward scheme: von Neumann analysis

Consider the forward in time centered in space numerical scheme for the heat equation
tD n
wjn+1 = wjn + wj+1 2wjn + wj1
n


2
.
x
Consider simple solution for this difference equation on the form of

wjn = n e2ijlh (15)

where i = 1 and l is an integer, which can be thought of as a discrete version of the Fourier
harmonics where the amplitude is known as the amplification factor.

Theorem 2 (von Neumann)

A numerical scheme for an evolution equation is stable if and only if the associated largest ampli-
fication factor satisfies
|| 1 + O(t).

We skip the details of the proof here but in the nutshell it is due to the fact that in the linear case
the largest eigenvalue of the difference scheme matches the amplification factor of von Neumann.

Inserting the expression of wjn in (15) into the forward scheme (12) yields

Dt  2ilh 2ilh
 Dt
=1+ 2
e 2 + e = 1 2 2 (1 cos(2lh))
h h
i.e
Dt 2 h2
|| 1 2 sin (hl) 1 t .
h2 2D
Thus, the difference scheme (12) is conditionally stable. For the example (13) with D = 1/16 and
h = 1/11 we have stability if and only if
16
t 0.06,
2 112
which explains the results in Figure 13. The condition of a time step being as small as h2 is very
bad news for this method especially if we want to integrate for a long period of time.

9
 5
x 10
6
exact
dt=0.02

6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
3
x 10
3
exact
dt=0.2

3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 1: Explicit scheme for the heat equation.10 Numerical solution (crosses) compared to the
exact solution (solid) for (13) at time t=4 with 11 spatial grid points. Top: t = 0.02, Bottom:
t = 0.2
1.3 Implicit scheme for the heat equation

Instead of approximating the derivatives for the heat equation at time tn using a forward finite def-
erence in time let us instead consider an approximation at time tn+1 and use a backward difference
in time. We arrive at the implicit/backward scheme for the heat equation

Dt  n+1 
wjn+1 = wjn + wj+1 2wj
n+1
+ w n+1
j1 . (16)
h2

Note that because wjn+1 is not given explicitly in terms of wjn , the time evolution of this scheme
necessitates the inversion of a linear system of equations:

(I A)X n+1 = X n + F n+1

where
2 1 (tn+1 )
w1n

1 2 1 0
wn
..

2 .. .. ..
X = . ,A = , F n+1 = D

. . . . .
.
n
1 2 1 0
wM
1 2 (tn+1 )
Dt
where D = h2 .

It is clear from its derivation that this new scheme is consistent and is first order in time and second
order in space.

Let us look at the stability properties of (16) using von Neumanns method. The amplification
factor in this case satisfies
Dt
= 1 4 2 sin2 (lh)
h
i.e,
1
0= Dt
1
1 + 4 h2 sin2 (lh)
independently on the values of D, t or h. The implicit scheme is unconditionally stable.

Because of this unconditional stability property the implicit scheme appears to be much superior
than its explicit counterpart because in principle it can be run with an arbitrarily large time step
and will still provide sensible results but it is much more expensive in terms of numerical operations
per time step. Moreover, because it is only first order accurate in time, the time step required to
achieve an accuracy on the order of h2 for a given spatial grid size h is t h2 , i.e, as small as
the time step required to achieve a numerical stability with the explicit scheme...Ideally, we want
a method which is both accurate and stable for large values of t at least as large as h. The
Crank-Nicholson scheme described below is one of such methods.

11
1.4 The Crank-Nicholson scheme

Crank-Nicholsons scheme combines the forward/explicit and the backward/implicit schemes in (12)
and (16) to provide a method which is both second order in time and space and unconditionally
stable. A straight average of the two schemes (12) and (16), yields the Crank-Nicholson scheme for
the heat equation
 n+1 
wjn+1 = wjn + wj+1 2wjn+1 + wj1
n+1 n
+ wj+1 2wjn + wj1
n
. (17)
2
Notice that the Crank-Nicholson scheme is implicit and as the backward-method it involves the
solution of a linear system at each iteration.
1 1 1
(I A)X n+1 = (I + A)X n + (F n+1 + F n ).
2 2 2

Exercise 1 Show that the Crank-Nicholson scheme is consistent to the second order in both time
and space and it is unconditionally stable. Hint: To show that it is second order accurate in both
time and space, consider Taylor expansion in both time and space for the solution u(x, t) about the
cell-center (xj , tn + t/2), highlighting the fact that the Crank-Nicholson method is centered in both
time and space.

2 Time splitting methods

Time splitting is an useful technique which consists on breaking down a complex PDE equation
into a few simple parts for which numerical schemes are easily constructed and analyzed. For
illustration we consider an advection diffusion equation in one space dimension, a reduced model
for the Focker-Plank equation
ut + a(x, t)ux = Duxx .

After discretization of the spatial derivatives, using the appropriate difference schemes or some
other technique to approximate the spatial derivatives, we arrive to a linear system of differential
equations with respect to time
d
w = Aw + Bw (18)
dt
where A is the discrete advection operator and B the discrete diffusion operator, and w(t) =
(wj (t))1jM (u(xj , t))1jM . Time splitting consists in dividing this linear systems onto two
natural systems that are integrated separately and successively during each time step, each corre-
sponding to the operators A and B, respectively. To integrate the discrete system (18) from t to
t + t, we proceed as follows.

The time splitting algorithm:

1. Let wjn = wj (tn ) be given at time tn

12
 d 1
2. Solve dt wj = Aw1 on [t, t + t], with w1 (t) = w(t)
d 2
3. Solve dt wj = Bw2 on [t, t + t], with w2 (t) = w1 (t + t)

4. Set w(t + t) = w2 (t + t), t = t + t and proceed to step 2.

The main advantage of the time splitting method is that it permits to use numerical schemes that
are known to converge and perhaps readily implemented for each one of the differential operators
separately, e.g. the advection operator for which various numerical schemes will be designed below
and the diffusion operator introduced above. However, the splitting methods introduce splitting
errors which limit the overall order of accuracy to first order, as it is revealed by the consistency
analysis performed next.

Let us analyze the consistency of the splitting methodology to see whether this is a sensitive method
to use in practice. Assume A and B are two linear time independent operators, as in the case of
the advection-diffusion problem. According to the theory of linear systems of differential equations,
the solution to the total linear system (18) is given by

s2
 
s(A+B) 2 3
w(t + s) = e w(t) = I + s(A + B) + (A + B) + O(s )I w(t)
2

s2 2
 
2 3
= I + s(A + B) + (A + B + AB + BA) + O(s )I w(t),
2
for s small (thinking s = t ). One step of the splitting scheme yields

s2 2
  
sB 2 sB sA 3 1 2 2 3
w(t+s) = e w (t) = e e w(t) = I + sB + B + O(s )I I + sA + s A + O(s )I w(t)
2 2

s2 2
 
2 3
= I + s(A + B) + (A + B + 2BA) + O(s )I w(t).
2
Thus w(t + t) w(t + t) = O(t3 ) if AB = BA, i.e, if A and B commute with each other and
w(t + t) w(t + t) = O(t2 ) if A and B do not commute, which is generally the case in practice.
Therefore the time splitting method is only first order accurate in time, but surprisingly it often
provides good results. Nevertheless, a more elaborate version of the time splitting method which
is second order accurate regardless the commutativity of the operators A and B is introduced by
Strang and, therefore, known as the Strang-splitting method. Stang splitting method consists in
symmetrizing the splitting operation by introducting an extra step to the time splitting algorithm,
where one of the two operators is solved twice with one half time step. The Strang-splliting method
is given next.

Strang-Splitting

1. Let wjn = wj (tn ) be given at time tn

d 1
2. Solve dt wj = Aw1 on [t, t + t/2], with w1 (t) = w(t)
d 2
3. Solve dt wj = Bw2 on [t, t + t], with w2 (t) = w1 (t + t/2)

13
 d 3
4. Solve dt wj = Aw3 on [t + /2, t + t], with w3 (t + t/2) = w2 (t + t)

5. Set w(t + t) = w3 (t + t), t = t + t and proceed to step 2.

Notice that in practice, when this algorithm is called successively in time, only the first and last
time steps need to involve half time steps: performing step 5 followed by step 2 is equivalent
to performing one full time step with the operator A. This may explain why the standard time
splitting performs very well in practice.

Exercise 2 Show that the Strang-splitting method is second order accurate.

3 Introduction to quasi-linear equations and scalar conservation

laws

A partial differential equation of the form

ut + a(x, t, u)ux + b(x, t, u) = 0, (19)

is said quasi-linear. If in addition the coefficient a is independent of u and b is linear in u, then the
equation is said linear. For practical purposes, more often than not, our computational domain
will be the finite interval 0 x 1, for which a boundary condition is needed at least at one end
of the domain. Also we assume that the equation in (19) is supplemented by an initial condition
at t = 0:
u(x, 0) = u0 (x). (20)
The couple pde + initial condition (19) and (20) is often referred to as a the Cauchy problem.

3.1 Prototype examples

Our main focus here is on the following two prototypes of equations: advection equations

ut + a(x, t)ux = 0 (21)

and scalar conservation laws

ut + (f (x, t, u))x = 0, (22)
both for being very important in many applications and as simple prototypes for more general and
more complex models used in applications. For instance, in the general area of fluid mechanics,
the advection equation is used as a model for the transport of tracers such as temperature, density,
or the concentration of a certain chemical by the fluid flow, when the latter does not change with
changes in the tracer u. It is widely used in biology and atmosphere-ocean sciences to model the
concentration of pollutants and other substances in the presence of wind and or ocean currents.
The term a(x, t) represents the flow velocity, i.e, the wind or the stream of watercalled as the

14
advection velocity or speed of propagation and u(x, t) is some measure of the tracercalled the
advected variable. More often advection models involve two to three space dimensions:

ut + a1 (x, y, z, t)ux + a2 (x, y, z, t)uy + a3 (x, y, z, t)uz = 0

or ut + a(x, y, z, t) u = 0
where a = (a1 , a2 , a3 ) represents the three dimensional velocity field and is the gradient operator.
Here we consider only the one space dimension case, extension to higher dimensions is mostly only
technical. In some applications, the advection coefficient is stochastic, i,e, depends on a random
variable or that the PDE/conservation law is forced by a stochastic forcing. Solving such a stochastic
model numerical can be easily handled by the techniques developed here.

The conservation law equation on the other hand, models the evolution of a conservative quantity.
That is a quantity whose variation inside a closed domain is equal to its flux across the boundaries,
i.e, the amount which flows in minus the amount which flows out, of the closed domain. In one
space dimension this can be stated as follows. Let u(x, t) be the concentration density of such a
conserved quantity within an interval [x, x + dx]. Let f (x, t, u) be the flux of u at the extremity x
and f (x + dx, t, u) the flux at x + dx. This is illustrated in Figure 2 for both the 1D and 2D cases.
R x+dx
By assuming the rate of change in the total quantity x u(, t, u)dy is equal to the total flux
through x and x + dx, we can write
x+dx

Z
u(x, t)dx = f (x + dx, t, u(x + dx, t)) + f (x, t, u(x, t)). (23)
t x

Notice, the minus sign in front of f (x+ dx, t, u) guarantees that the flux on the right side is directed
inward if f is positive and outward if f is negative. The opposite is true for the flux on the left side.
The equation (23) is known as the integral form of the conservation law. Notice that no regularity
with respect to x for u or f is required when we derived this equation. As we will see later this has
some important consequences in designing numerical schemes for solving such equations.

Dividing by dx both sides of (23) and letting dx go to zero yields the differential form of the
conservation law in (22), provided both u and f are smooth. However, in many textbooks and
research papers, only the differential notation is used even when u or f are not smooth, in this
case u is no longer a solution in the classical sense but only in a weak sense, which will be clarified
below.

Another way to derive (22) from (23) is by noting that

Z x+dx
f (x + dx, t, u(x + dx, t)) + f (x, t, u(x, t)) = (f (y, t, u))x dy,
x

which leads to the equality of the integrands because the length dx is arbitrary. This is actually
the way the conservation law generalizes to 2 space dimensions and more. In fact, consider a closed
domain of R2 and let be its boundary. Then equating the total rate of change of u in to
the total flux of u across the boundary of yields

Z Z
u(x, y, t)dxdy = F (x, y, t, u) nd
t

15
 Flux in
Flux out

X X + dX

Flux out

Flux in

Flux out

Flux in

Figure 2: The integral form of the conservation law states that the total rate of change in u is
compensated by the flux-in minus the flux-out.

where F is the flux vector and n is the unit normal vector to which is directed outside .
Invoking the divergence theorem yields

Z Z
u(x, y, t)dxdy = F (x, y, t, u) dxdy
t
for all bounded domain . This yields the conservation law in differential form
ut + F (x, y, t, u) = 0.

Links between the advection and the conservation law equations

The advection and the conservation law equations are intimately interconnected. In the case when
f = f (u) the conservation equation (22) can be rewritten in the advective form as
df (u)
ut + ux = 0
du
and when a is constant with respect to x the advection equation can also be viewed as a conservation
law. Moreover, note that in general every quasi-linear equation can be written as a conserved part
plus a forcing
ut + (f (t, x, u))x + c(x, t, u) = 0.
Such equations are sometimes called balance laws and they are widely used in practice. The search
for adequatenumerically well balanced schemes for this kind of equations has recently became a
very active research area.

16
Finally, we note the advection equation in (21) is linear while the conservation equation in (22) can
be non-linear if u f 6= 0. A very commonly used prototype example of a nonlinear conservation
law is the celebrated Burgers equation:
1
ut + (u2 )x = 0, (24)
2
which becomes
ut + u ux = 0,
when written in advective form.

3.2 Solutions by the method of characteristics

Consider the quasi-linear equation in (19). Let x = x(t) be a parametric curve in the (x, t) plane
such that x = a(x, t, u(x(t), t)) where u(x(t), t) z(t) is the solution to (19) along this curve. Using
the chain rule and plugging into the equation in (19) yields

u u u u
z = + x = + a(x, t, u) = b(x, t, u) = b(x, t, z).
t x t x
i.e, finding a solution for the quasi-linear equation reduces to solving the following system of first
order ordinary differential equations.

x = a(x, t, z) (25)
z = b(x, t, z)
x(0) = x0 , z0 = u(x0 , 0) = u0 (x0 )

Equations (25) are known as the characteristic equations and the resulting solution curves x =
x(t), x(0) = x0 are called characteristic curves.

Example 1: Solution to the advection equation

For simplicity we assume that the advection speed a is constant, in which case the advection
equation reduces to since
ut + aux = 0.
The characteristic equations for this simple example are

x = a
z = 0

whose solution is x(t) = x0 + at, z(t) u(x(t), t) = u0 (x0 ). Two key important points should be
noted here. i) The characteristic curves are straight lines. ii) The solution u is constant along the
characteristic lines. The characteristic curves for the advection equation are sketched on Figure 3
for both a > 0 and a < 0. Note that when a > 0 the characteristics are directed to the right and
when a < 0 they are directed to the left. In some sense the sign of a indicates the direction of
propagation of information. In fact the advection equation is also called the one-way wave equation,
where a is the speed of propagation of the wave.

17
 Time, t

X=

at
X0

+
X0
+a

X=
t

X0 X0 X
a<0 a>0

Figure 3: Characteristic lines for the advection equation. When a > 0 the characteristics are
directed to the right and when a < 0 they are directed to the left.

To find the solution u(x, t) at an arbitrary point (x, t) in the x-t plane, one needs to follow the
characteristic line passing through (x, t) back to its original point at t = 0: (x0 , 0) with x = x0 + at.
This leads to
u(x, t) = u0 (x0 ) = u0 (x at). (26)

Example 2: Burgers equation

The Burger equation constitutes a somewhat more complex example a quasi-linear PDE. However,
we still can, in principle, construct exact solutions using the method of characteristics.

The system of characteristic equations for Burgers equation is given by

x = z
z = 0.

The characteristic solution is thus given by

u(x(t), t) = u0 (x0 ); where x(t) = x0 + u0 (x0 )t. (27)
Again note that the characteristics are straight lines and the solution is constant along the char-
acteristic lines, with one important difference, however; the characteristic curves are no longer
parallel to each other. As we will see below this has rather unpleasant consequences. Provided
the characteristics lines do not cross each other, which is guaranteed for at least a short period of
time if the initial data u0 is continuous, the solution to Burgers equation is given by the following
implicit formula
u(x, t) = u0 (x u0 (x0 )t), x = x0 + u0 (x0 )t.
The characteristic lines associated with Burgers equation are illustrated in Figure 4.

Exercise 3 Use the method of characteristics to solve the following quasi-linear equations.
ut + xux = 0

18
Time, t t
X0)
u0(
+
X0
X=
X=
X0
+a
t

X0 X0 X
u0(x) > 0 u0(X) < 0

Figure 4: Characteristic lines for Burgers equation. When u0 (x) > 0 the characteristics are
divergent and when u0 (x) < 0 they are converging toward each other.

and
ut + ux + x = 0.
Write down the solution u(x, t) and draw the characteristic curves.

3.3 Notion of shocks and weak solutions

Note that because the slope of the characteristic curves x = x0 + u0 (x0 )t for Burgers equation (4)
increases when u0 (x0 ) increases and decreases when u0 (x0 ) decreases, the characteristic curves will
accordingly diverge or converge toward each other (see Figure 4). Two convergent characteristic
lines will ultimately cross each other at some point in the x-t plane. Beyond such intersection point
the characteristic solution is no-longer valid, because the value of u(x, t) at such a point is not
univaluedone can follow back either one of the two intersecting characteristic lines.

One way to correct for this flaw is by stopping the characteristic lines as soon as they cross each
other. Let be the set of such crossing points in the x-t plane. The solution can then be defined
on both sides of by following the corresponding characteristic line back to its origin. Below we
will see that is a parametric curve on the form x = s(t) as show on Figure 5 and constitutes a
curve of discontinuity for u(x, t). Such a curve is called a shock by analogy to gas dynamics. One
of the main difficulties in practice is to find the shock curve x = s(t). For any given (x1 , t1 ) one
has to determine whether two characteristic lines cross each other prior to time t1 along the curve
x = x1 .

Exercise 4 Show that a shock forms in the solution for Burgers equation if and only if the initial
condition satisfies
u0 (x) < 0 for some x

19

Shock curve

Time, t

Figure 5: The shock curve separates two regions of the x-t plane where the solution is smooth
and is uniquely determined by the characteristics. The solution is discontinuous across the shock
curve.

and that the first time a shock occurs is given by

1
T = .
minx u0 (x)

After a shock is formed the solution u(x, t) is no longer valid in the classical sense except for its
restrictions on the sub-domains located on either side on the shock. Nevertheless, such solution
can be defined in the weak sense on the whole x-t plane.

Definition 3 A function u(x, t) is said to be a weak solution for the conservation law
ut + (f (x, t, u))x = 0
if for any test function (x, t) sufficiently smooth (e.g. C 1 ) with a compact support1 in (, +)
(0, +), the solution u(x, t) satisfies
Z + Z + Z + Z +
u(x, t)t (x, t) dxdt + f (x, t, u)x (x, t) dxdt = 0. (28)
0 0

Remark:
Note that according to the definition of weak solutions, given above, a C 1 function u(x, t) is a
1
i.e, there exist a bounded rectangle [t1 , t2 ] [a, b] (, +) (0, +) such that phi(x, t) = 0 out side this
rectangle.

20
solution to the conservation law in the classical sense if and only if it is a solution in the weak
sense. Therefore the notion of weak solutions is more general and the set of weak solutions contains
discontinuous solutions as well as the classical C 1 solutions as a special subset. However, in some
situations weak solutions are not unique, in the sense that one initial value problem can have more
than one weak solution. Selecting the physically relevant solution can be tricky. For Burgers
equation for example, the physical solution coincides with the vanishing viscosity solution:

u(x, t) = lim u (x, t)

0

where
u u 2 u
+ u = 2,
t x x
under the grounds that the unviscid Burger equation is a mathematical idealization of the viscous
Burger equation when the viscosity is very small. However, given two weak solutions it is not easy
to identify which one is the limiting viscosity solution and which one is not. The answer is provided
by an extra condition known as the entropy condition satisfied by the physical solution. We will
see this in detail below.

Note that the notion of weak solutions is very abstract and it is not obvious how one would
handle this in practice. Nevertheless the theorem below provides the necessary ingredients both for
constructing weak solutions and to gain physical insight.

Theorem 3 (Rankine-Hugoniot Condition) Let be a curve in (, +)(0, +) parametrized

by x = s(t). Let u(x, t) be a C 1 function on both side of but possibly not defined and discontinuous
across the curve . Assume u is a solution to the conservation law

ut + (f (u))x = 0

on all points (x, t) not on . For each point (x1 , t1 ) we set

u (x1 , t1 ) = lim u(x, t).

1 2 (x,t)(x1 ,t1 )

i.e. the limits from the right and from the left of . Then u(x, t) is a weak solution for the
conservation law if and only if the shock speed s satisfies

ds f (u+ ) f (u )
s = . (29)
dt u+ u

The proof of this theorem is not terribly hard but it is quite technical and therefore left as an
exercise for the interested student.

3.4 Discontinuous initial data and the Riemann problem

As it is pointed out above, the notion of weak solutions permits to define discontinuous solutions
for conservation laws. Here we propose to construct such weak solutions with discontinuous initial

21
data. For simplicity, we consider the Burger equation with discontinuous initial data consisting of
two constant states, a left and a right state:

ut + uux = 0

uL if x < 0
u0 (x) = (30)
uR if x > 0.

The problem in (30) is known as the Riemann problem.

We propose to construct simple weak solutions for the Riemann problem associated with Burgers
equation, using the Rankine-Huguniot condition (29).

Shock waves:
Consider a discontinuous function which consists of the two left and right constant states on both
sides of a shock curve : x = s(t); s(0) = 0

u(x, t) = uR if x > s(t) (31)

u(x, t) = uL if x < s(t). (32)

According to the Rankine-Hugoniot condition we have

1 u2R u2L uR + uL
s = = .
2 uR uL 2
Note that the shock speed in this case is constant and the curve is a straight line. Also recall
that the speed of the characteristic lines on each side of the shock is simply uL and uR , respectively.
Therefore the speed of the shock is exactly halfway between the left and right characteristic speeds.
This has physical sense and the associated weak solution is called a shock wave. A rough sketch of
the shock wave solution is given in Figure 6 for both cases when uR < uL and when uL > uR . Note
that in the first case the characteristics run into the shock line and stop and therefore the solution
on both sides of the shock is consistent with the characteristic solution while in the second case
the characteristics diverge away and the region surrounding the shock is not reached by any of the
characteristics.

Rarefaction waves:
Now assume that uL < uR so the characteristics emanating from both side of the discontinuity
are divergent from each other. In this case we can actually construct another weak solution to the
Riemann problem associated with the Burgers equation. First note that for t > 0 the function
u(x, t) = x/t, satisfies Burgers equation

ut + uux = 0.

For t > 0 consider

uL if x < tuL
u(x, t) = x/t if tuL < x < tuR (33)
uR if tuR < x.

First note that u(x, t) is a solution to Burgers equation on each one of the designated parts of
the domain and is continuous on the x-t plane. We can therefore show by simple integration by

22
 time T

: x = (ur+ul)/2 t
111111111111111111111111
000000000000000000000000
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
u = ul
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
u = ur
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111

X
time T

: x = (ur+ul)/2 t
000000000000000000000000
111111111111111111111111
111111111111111111111111
000000000000000000000000
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
u = ul
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
u = ur
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111

X
Figure 6: Shock wave solutions for Burgers equation. The two cases when uL > uR (top) and
when uL < uR (bottom) are shown.

23
 time T

Rarefaction fan
u = x/t
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
1111111111111111111111111
0000000000000000000000000
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
u = ul 0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
u = ur
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111
0000000000000000000000000
1111111111111111111111111

X
Figure 7: Rarefaction wave solution for Burgers equation. The rarefaction fan is shown.

parts that indeed u(x, t) is a weak solution (see exercise 5 below). This type of solution is called a
rarefaction wave by analogy to compressible gas dynamics and the solution u = x/t in the middle
is referred to as a rarefaction fan, see the illustration in Figure 7.

In summary, this shows that the Riemann problem for Burgers equation has at least two weak
solutions when uL < uR , one is a shock wave and the other is a rarefaction wave. Therefore, weak
solutions for conservation laws are in general non-unique.

Exercise 5 Let be a bounded open set in the xt-plane. Let be a curve passing through
dividing it onto two disjoint open subsets 1,2 such that = 1 2 . Let (x, t) be a smooth
function (e.g. C 1 ) supported in , that is vanishes outside a compact set K . Let

u1 (x, t) if (x, t) 1
u(x, t) =
u2 (x, t) if (x, t) 2 ,
where u1 , u2 are two C 1 functions satisfying the Burger equation in 1 , 2 , respectively. Use inte-
gration by parts to show that if in addition u is continuous across , then
1
Z
ut + u2 x dxdt = 0.
2
Deduce that (33) is a weak solution to Burgers equation.

3.5 Non-uniqueness of weak solutions and the entropy condition

As illustrated above with the example of a Riemann problem for Burgers equation, weak solutions
are non-unique. However, common sense suggests that for any given Cauchy problem, only one
solution is physically relevant. We need an additional constraint to choose this physically relevant
solution among all the weak solutions. In fact, in reality some viscosity is always associated with
a given conservation law, so instead we have
2 u
ut + (f (u ))x = , (34)
x2

24
and the zero viscosity limit, 0, is just a convenient mathematical idealization. It is true in
many physical applications! Therefore, one universally accepted criterion states that the physically
relevant weak solution for the conservation law ut + (f (u))x = 0 is the limit of the solution to the
viscous equation (34) when 0. On the other hand it is easy to show that, when combined
with appropriate initial conditions, the latter has a unique solution. In practice, however, it is
not clear how to establish if a given weak solution is actually the vanishing viscosity limit or not.
The answer to this question is provided by the concept of entropic solutions. In a nutshell, the
entropy condition states that the physical solution satisfies a general principle of thermodynamics
that the entropy always decreases. It remains to find which among the weak solutions for a given
conservation law satisfies the so-called entropy condition. There are many versions of the entropy
condition for a given conservation law

ut + (f (u))x = 0.

i) A somewhat abstract version of the entropy condition, but with a clear physical significance
is as follows. Given a conservation law

ut + f (u)x = 0.

A convex function (u) and a flux function (u) are called an entropy/entropy flux pair if

(u) = (u)f (u).

Given an entropy/entropy flux pair (, ), a solution u for the conservation law is said to be
an entropic solution if it satisfies
(u) (u)
+ 0, (35)
t x
in the weak sense. In many physical problem the entropy function (u) is some measure of
energy. It can be shown that for Burgers equation, an entropy/entropy flux pair is given by
2
(u) = u2 ; (u) = u3 .
3
Here (u) = u2 can be thought of as an energy density and (u) is the energy flux. The most
apparent merit of this formulation is that it generalizes easily to systems of conservation
laws.

ii) A more practical version of the entropy condition is the following:

 
1
u(x + z, t) u(x, t) C 1 + z, z, t > 0. (36)
t

iii) Perhaps the most abstract of them is due to Kruzkov. A (weak) solution to the conservation
law is said an entropy solution in the sense of Kruzkov if, in addition, it satisfies
Z Z +
sign(u k) [(u k)t + (f (u) f (k))x ] dxdt 0 (37)
0

for all real constants k and all test functions 0. This version of the entropic solution
is useful for theory. It is mainly used by mathematicians to prove existence and uniqueness
theorems.

25
 iv) Finally, we give the Lax entropy condition for shocks: A shock solution for the Riemann
problem with left and right states uL , uR is said an entropic shock if the shock speed, s,
satisfies
f (uR ) < s < f (uL ). (38)

It is easy to see that only one of the two shock solutions for the Riemann problem for Burgers
equation is an entropic solution. Namely, the one associated with the case uR < uL (by the Lax
entropy condition iv) as well as condition ii)). Moreover, it can be shown that the unique entropic
solution in the case uL < uR is the rarefaction wave in (33).

4 Finite difference schemes for the advection equation

We start by discussing some basic simple finite difference schemes for the advection equation

ut + aux = 0,

where a is a positive constant. In principle these schemes are easily generalized to non-constant
advection speeds with an arbitrary sign.

4.1 Some simple basic schemes

Throughout this paper we will assume the following discretization of space-time domain

xj = jx; tn = nt,

where x, t > 0 are respectively the spatial and time step sizes. We denote by unj the approxi-
mate/numerical solution to the solution u(xj , tn ).

Perhaps the most obvious scheme to attempt for the advection equation, which unfortunately turns
out to be unstable, is obtained by taking a first order forward finite differencing in time and a centred
differencing in space:
at n
un+1 = unj (u unj1 ). (39)
j
2x j+1
This scheme is referred to below, simply, as the centred scheme. Other simple possibilities are to
take a first order derivative in space either to the left or to the right, combined with the forward
differencing in time, yielding the so-called first order upwind and downwind schemes, respectively:
at n
un+1 = unj (u unj1 ) (40)
j
x j
and
at n
un+1 = unj (u unj ). (41)
j
x j+1
Note that those two schemes are also known as the upstream and downstream schemes; depending
on the application: hydrodynamics or gas dynamics (ocean v.s. atmosphere). The word upwind

26
 n+1 n+1 n+1 n+1
(a) (b) (c) (d)

n n n
j1 j j+1 j1 j j j+1 j1 j+1
n1

Figure 8: Simple finite difference stencils for the advection equation: (a) forward in time centred
in space, (b) upwind, (c) downwind, (d) leap-frog.

refers to the fact that the finite differencing is performed in the direction opposite to the wind and
downwind when the difference scheme follows the wind direction. Accordingly, when a < 0 the
scheme in (41) becomes upwind and the scheme (40) becomes downwind.

Warning: As we will see below both the centred and the downwind schemes (39) and (41) are not
recommended in practice because they are both unstable.

A slightly more sophisticated scheme is the leap frog scheme, which uses centred differences in both
space and time:
at n
un+1 = ujn1 (u unj1 ). (42)
j
x j+1

The stencils for the four schemes listed above are given in Figure 8.

4.2 Accuracy and consistency

Definition 4 Let Lh (uh ) = 0 denote the numerical discretization using a certain method for a
given partial differential equation denoted by L(u(x, t)) = 0, with a time step, t, and grid spacing,
x. The numerical scheme is said to be consistent if the truncation error:

h = L(u) Lh (u) (43)

satisfies
lim h = 0.
t,x0

The scheme is said consistent of order (p, q) accurate or simply of order (p, q) if

h = O((x)p + (t)q ).

Examples:

Consider the advection equation

L(u) ut + aux = 0.

27
Using simple Taylor expansions, it is easy to see that the centred scheme (39) is consistent of
order (2,1), namely

u(x, t + t) u(x, t) u(x + x, t) u(x x, t) t (x)2

ut + aux a = utt (x, ) a uxxx (, t)
t 2x 2 6
= O(t + (x)2 ).

i.e, this scheme is first order accurate in time and second order in space while the upwind and
downwind schemes (40) and (41) are only first order in both space and time,

h (upwind/downwind) = O(t + x).

The leap-frog scheme (42), on the other hand is second order in both space and time

h (leapf rog) = O((t)2 + (x)2 ).

Exercise 6 Show that the leap-frog scheme (42) is second order accurate in both time and space.

4.3 Stability and convergence: the CFL condition and Lax-equivalence theorem

Definition 5 A numerical scheme for an evolution equation on a finite interval [0, T ], on the form

un+1
j = Sh (unj )

is said stable if there exists a constant C > 0 such that

||un || C||u0 ||

for a certain norm ||.|| in RN , where N is the number of spatial grid points.

Convergence of the numerical scheme

The convergence of a numerical solution uh , obtained by a given numerical scheme, to the solution
u of the original continuous equation is solved by the celebrated Lax-equivalence theorem.

Theorem 4 (Lax-equivalence theorem) The approximate numerical solution to a well posed

linear problem converges to the solution of the continuous equation if and only if the numerical
scheme is linear, consistent, and stable. The rate of convergence and the order of accuracy of the
numerical solution is equal to the truncation error of the numerical scheme.

This elegant and powerful theorem is often summarized as follows

consistency + stability = convergence .

28
von Neumann stability analysis
The study of stability of a numerical scheme can be very tedious but the use of Fourier analysis
when appropriate simplifies it a great deal. This idea was first used by von Neumannapparently.
For simplicity we assume that any discrete function, fj , i.e, defined on the grid points xj = jx
by its values fj = f (xj ) can be expanded in discrete Fourier modes
N/2
X
fj = l eijx2l + complex conjugate terms
l=0

where i = 1 and l are complex Fourier coefficients. N here is the number of spatial grid points
and N/2 is known as the Nyquist number. It represents the largest wavenumber represented on an
N-points grid.

To simplify the notation we set

l = 2lx.
For the numerical solution unj evolving in the discrete time, tn , we have

N/2
X
unj = nl eijl .
l=0

Note that the upper script n is an index not a power.

Theorem 5 (von Neumann Stability) A numerical scheme for an evolution PDE is stable (in
the sense of von Neumann) if and only if the ratio

n+1
l
l =
nl
known as the amplification factor, of discrete Fourier coefficients of the numerical solution, satisfies

|l | (1 + t), l = 0, , N/2.

Although the proof of this theorem is almost trivial, by Parsevals equality, it has a huge significance
and a big impact on our way of studying numerical methods, because it is very easy to use, especially
for linear problems. In fact, when the numerical scheme is linear, it is enough to consider solutions
on the form of a single Fourier mode
unj = nl eijl . (44)

Plugging in (44) into the

the centred scheme (39) yields

 
n+1 eijl = n eijl n eijl eil eil
2

29
 n+1
where = at/x. Thus the amplification factor, n , is given by

= 1 i sin(l ).

Thus, ||2 = 1 + 2 sin2 (l ) > 1 + t for some values of l , for all t sufficiently small,
therefore the centred scheme (39) is unstable as suggested above.

the backward (in space) first-order scheme (40) yields

= 1 (1 eil ) = 1 (1 cos(l )) i sin(l ),

||2 = 1 + 2 (1 + cos2 (l ) 2 cos(l )) 2(1 cos(l ) + 2 sin2 (l )
= 1 + 22 + 2(1 ) cos(l )) 2.

Clearly if < 0, i.e a < 0, > 1 + t for some values of l , for all t sufficiently small,
and when > 0 (a > 0), 1 provided 0 1. In other words, the upwind scheme
is conditionally stable: t x/|a| or |a|t x while the downwind scheme is always
unstable. Similarly we can show that the forward-scheme (41) is stable if 1 0 and
unstable otherwise. Notice that the upwind scheme amounts to taking the derivative in the
direction opposite to the advection speed, i.e, toward the direction where the information
came from.

the leap-frog scheme (42) yields

2 = 1 2i sin(l ),

which has 2 roots q

= i sin(l ) 1 2 sin2 (l ) if || < 1 (45)

| |2 = 1 if || < 1.
When || > 1, one can always find a value for l for which the two roots are both imaginary
and then necessarily one of them has to be strictly larger than one for all values of t. Thus, as
the upwind scheme, the leapfrog scheme is conditionally stable: t x/|a|. Note however,
that unlike the previous schemes, von Neumann analysis for the leap-frog method leads to two
amplitude modes, , for a given spatial mode exp(ijl ). Nevertheless, as we will see below,
only one of them is physical, namely, + , that is it represents an approximation (converges)
to the exact solution, while the other one is an artifact of the numerical discretization. The
latter is often called a computational or a parasitic mode.

CFL condition:
The CFL condition, named after its discoverers, Courant, Friedrichs, and Lewy, states that if a
difference scheme for an evolution equation is stable then the domain of dependence, corresponding
to one time step, of the numerical scheme contains the domain of the dependence of the original
continuous equation, when both the time step and the spatial grid size t, x 0.

Lets first clarify the notion of domain of dependence. Given the partial differential equation,

ut = F (u, ux )

30
 Table 1: Some simple schemes for the advection equation and their properties.
Scheme Order of accuracy Stability CFL condition
Centred O(t + (x)2 ) Unstable Satisfied if t x/|a|
Upwind O(t + x) Stable if t x/|a| Satisfied if t x/|a|
Downwind O(t + x) Unstable Not satisfied
Leap-frog O((t)2 + (x)2 ) Stable if t x/|a| Satisfied if t x/|a|

with and initial condition

u(x, t) = u0 (x),
the domain of dependence of this equation at some point (x, t) is the set of all points x0 such that
u(x, t) depends on the value of u0 at x0 ). For instance according to the solution by the method of
characteristics to the advection equation (26), the domain of dependence of the advection equation
is reduced to one point
D(x,t) = {x0 = x at} .
The domain of dependence corresponding to one time step is

D(x,t+t) = {(x at, t)} .

This is illustrated in figure 9 for a > 0. The numerical domain of dependence of the numerical
schemes considered so far are as follows

Dcentred (x, t + t) = {(x x, t), (x + x, t)}

Dbackward (x, t + t) = {(x x, t), (x, t)}

Df orward (x, t + t) = {(x, t), ((x + x, t)}
Dleapf rog (x, t + t) = {(x x, t), (x, t t), (x + x, t)}
Note that, as shown in Figure 9, the point (xax, t) is within the range the domain of dependence
of all the above schemes so that the CFL condition is satisfied, provided the stability condition
|t x/|a| is satisfied, except for the forward scheme, corresponding to downwind differencing in
this case, which explains its instability. Notice that this makes physical sense. As it is suggested by
the method of characteristics, when a > 0 the solution to the advection equation uses information
on the left to advance forward in time while the forward scheme uses information on the right. Note
that the upwind and the leapfrog schemes do not have this problem. Also, the condition |a|t x
can be interpreted as a requirement that the numerical speed of propagation of information being
smaller than the advection speed of the continuous problem.

Interestingly note that the centred scheme (39) satisfies the CFL condition, if t x/|a|, as for
the leap-frog scheme, but is not stable. This is a good example illustrating the important fact that
the CFL condition is a necessary condition for stability but it is not sufficient.

Table 1 summarizes the properties of the four simple schemes we have covered so far.

31
 (X,t+Dt)
t+Dt

t
XaDt

X Dx X X +Dx

Figure 9: Domain of dependence of the advection equation for a > 0 and the CFL condition.

4.4 More on the leap-frog scheme: the parasitic mode and the Robert-Asselin
filter

According to table 1, the best we have, among the simple schemes seen so far, is the leap-frog
scheme; it is stable and second order accurate in both space and time. Notice also as such it is
relatively cheap and easy to implement. This may explain in part why this scheme is so popular in
the engineering and atmosphere/ocean communities. However, it has at least two drawbacks: 1)
it is a multistep (2 steps or 3 levels) scheme which means it needs the knowledge of the solution
at two successive time steps in order to advance to the next one and 2) it carries a parasitic mode
which may ruin the numerical solution when used for long time integrations, if it is not filtered
carefully. More details on the leap-frog scheme and its parasitic mode can be found in the literature
(e.g. Durran). Here we briefly illustrate and demonstrate the behaviour of this parasitic mode and
give a strategy for controlling it in practice.

Recall the von Neumann amplification factors associated with the leap-frog scheme
p
= i 1 2

where we set = sin(l ). Since | | = 1 we have

 
i
= e , = arctan
1 2
with 0 = arcsin() we have
+ = 0 , = + 0 ,
+ = ei0 , = ei(+0 ) ,

32
and the Fourier mode solution for the leap-frog scheme is given by

unj = (c1 n+ + c2 n )ejil = c1 ei(jl n0 ) + c2 (1)n ei(jl +n0 )

c1 e2li(xj atn ) + c2 (1)n e2li(xj +atn ) (46)

where we used the fact that = at/x, l = 2lx, nt = tn ; jx = xj and the approximation

arcsin() ; sin(l )/l 1.

The expression in (46) clarifies that the term representing + has the form f (x at) and therefore
provides an approximation for the solution to the advection equation while the remaining term
represents a wave moving in the opposite direction (to the left) and its amplitude oscillates between
positive and negative values. Clearly, the latter is an artifact of the numerical discretization, called
the computational or parasitic mode, and may lead to serious damage to the solution if it is not
controlled in some way.

There are many ways how to control the computational mode. One of them is to make sure that
the extra initial condition, i.e, solution at first time step, needed to advance the leap-frog method
is chosen so that the coefficient c2 , of the parasitic mode, is zero fin the decomposition of u0 into
Fourier modes. One easy way to guarantee that initially the parasitic mode is zero is to set as
second initial data at t = t, required to evolve the multi-step leap-frog method, to be
X
u1j = + 0 ijl
l l e
l

given at t = 0 we have X
u0j = 0l eijl .
l
However, we known that for long integration periods the parasitic mode can be excited and grow
just from round-off errors.

A safe commonly used strategy for controlling the computational mode for the leap-frog scheme,
known as the Robert-Asselin filter, is given next. The Robert-Asselin filter consists in averaging
the solution at every time step by using the previous and the future solution, at t t and t + t,
respectively, by introducing an extra-filtering step. Let unj denote the solution filtered in such a
way. The two step leap-frog plus Robert-Asselin filtering scheme is given by

un+1
j = ujn1 (unj+1 unj1 )
unj = unj + (un+1
j 2unj + ujn1 ), (47)

where is a small filtering parameter usually taking to be = 0.06. Some atmospheric models used
for cloud physics use values as large as = 0.3 (Durran). It is important to note that the filtering
step destroys the second order accuracy in time and the resulting scheme is only first order.

4.5 The Lax-Friedrichs scheme

The Lax-Friedrichs scheme is a clever modification for the unstable centred scheme (39), which
makes it stable. It consists of replacing the term unj by the average from the neighbouring cells to

33
obtain
unj+1 + unj1 at n
un+1 uj+1 unj1 .


j = (48)
2 2x
This is the celebrated Lax-Friedrichs scheme. As the centred scheme the Lax-Friedrichs scheme
(48) is second order accurate in space and first order in time. However, it is stable under the CFL
condition || |a|t/x < 1 . In fact, the associated von Neumann amplification factor is given
by
= cos(l ) i sin(l )
= ||2 = cos(l )2 + 2 sin(l )2 1 if || 1.

Exercise 7 Show that the Lax-Friedrichs scheme is first order in time and second order in space
and that its amplification factor is given by

= cos(l ) i sin(l ).

The apparent advantages of the Lax-Friedrichs scheme, when compared to the three other stable
schemes listed above is that it is second order in space and is a one step method, therefore easier to
implement in practice compared to the leap-frog method. However, it is only first order accurate
in time, which limits its use in practical applications and it is extremely dissipative as we will see
below.

4.6 Second order schemes: the Lax-Wendroff scheme

Perhaps the simplest way to achieve a one-step (2 levels) scheme, which is second order in both
time and space for the advection equation, is to resort to Taylor expansion in the time variable

(t)2
u(x, t + t) = u(x, t) + tut (x, t) + utt (x, t) +
2
using the advection equation ut = aux yields

a2 (t)2
u(x, t + t) = u(x, t) atux (x, t) + uxx (x, t) +
2
Now, we use second order centred differencing to approximate the spatial derivatives ux and uxx
to obtain the Lax-Wendroff scheme
n
2 n
un+1 = unj uj+1 unj1 + uj+1 2unj + unj1 .


j (49)
2 2

It is easy to show that the Lax-Wendroff scheme (49) is second order accurate in both space and
time and it is stable under the CFL condition at/x 1. The amplification factor for this
scheme is
= 1 i sin(l ) 2 (1 cos(l )).

34
Exercise 8 Show that the Lax-Wendroff scheme is second order in both time and space and it is
stable under the CFL condition at/x 1.

If instead we use a second order approximation of the first order space derivative in the Taylor
expansion using points that are only in the upwind direction, namely, xj2 , xj1 , xj when a > 0,
we obtain the following scheme, known as the Beam-Warmings scheme:

2 n
un+1 = unj 3unj 4unj1 + unj2 + uj 2unj1 + unj2 .


j (50)
2 2
This scheme can be derived by using polynomial interpolation. In fact, let
uj1 uj2 uj 2uj1 + uj2
P2 (x) = uj2 + (x xj2 ) + (x xj1 )(x xj2 )
x 2(x)2
be the 2nd degree polynomial interpolating u on the grid points xj , xj1 , xj2 . Using this poly-
nomial to approximate the derivatives ux , uxx at x = xj yields the Beam-Warming scheme. The
details are left as an exercise for the student.

Note that because the stencil of the Beam-Warming scheme uses two grid cells on the left of xj ,
the associated CFL condition becomes 0 at/x < 2. In fact we can show that this scheme
is stable under this CFL condition. Compared to Lax-Wendroffs scheme this new scheme allows
larger time steps. However, if the advection speed is not too large the time step might be restricted
to that of the Lax-Wendroff for accuracy reasons.

Exercise 9 Show that the Beam-Warming scheme is second order in both time and space and it
is stable under the CFL condition at/x 2.

Exercise 10 Derive the version of the Beam-Warming scheme when a < 0.

4.7 Some numerical experiments

Here we assess the performance of each one of the (stable) methods listed above, for the advection
equation, the interval [0, 1] with periodic boundary conditions and two different initial data. We
consider a smooth initial condition consistent of a hump-like shaped profile,

u0 (x) = exp(100(x .5)2 ),

and a non-smoothpiece-wise constant, initial condition, called here a square wave

u0 (x) = 0 if |x 0.5| > 0.25, u0 (x) = 1 if |x 0.5| < 0.25.

The advection velocity is assumed constant and normalized to a = 1 and integrate to time t = 1
so that the initial profile is moved forward by a distance equal to the length of the interval [0, 1]
and by periodicity we have u(x, t = 1) = u0 (x). We use a mesh size x = 0.01 and a time step
t = 0.008 corresponding to a Courant number = t/(|a|x) = 0.8.

35
 1 1
Exact Exact
upwind Leap frog
0.8 LaxFriedrichs 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0.2 0 0.2 0.4 0.6 0.8 1 1.2 0.2 0 0.2 0.4 0.6 0.8 1 1.2

1.2 1.2
Exact Exact
1 LaxWondroff 1 BeamWarming

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

0.2 0.2
0.2 0 0.2 0.4 0.6 0.8 1 1.2 0.2 0 0.2 0.4 0.6 0.8 1 1.2

Figure 10: Solution to the advection equation using different finite difference schemes. Case of a
smooth hump advected periodically through the interval [0, 1]

36
 Exact Exact
1.5 upwind 1.5
Leap frog
LaxFriedrichs RA Leap frog
1 1

0.5 0.5

0 0

0.5 0.5
0.5 0 0.5 1 0.5 0 0.5 1

Exact
1.5 1.5 Exact
LaxWondroff
BeamWarming

1 1

0.5 0.5

0 0

0.5 0.5
0.5 0 0.5 1 0.5 0 0.5 1

Figure 11: Same as Fig. 10, except for the non-smooth, piece-wise constant, square wave. Note that
for this case results with both the plain leap-frog and leap-frog combined with the Robert-Asselin
filter are shown.

37
In Figure 10, we plot the numerical solutions, obtained with each one of the 5 methods, against the
exact solution for the advection equation. In Figure 11 we report the similar plots corresponding
to the non-smooth square wave. Here we note a few important points.

First, for the smooth solution in Figure 10, as expected the second order methods (leapfrog,
Lax-Wendroff, and Beam-Warming) are highly accurate whereas the first order upwind and
Lax-Friedrichs methods yield very unsatisfactory results. They both suffer from an excessive
dissipation; that is the wave amplitude decays in time. Surprisingly, the upwind scheme seem
to perform better than Lax-Friedrichs, although, the latter is second order accurate in space.
Second, for the non-smooth case in Figure 11, on the other hand, the two first-order methods
seem to perform better than the three second order methods, although they tend to smooth
out the discontinuity. The second order schemes exhibit high oscillations near the disconti-
nuities. This is typical for high order methods, they are exhibit an oscillatory behaviour near
shocks. Note also that Lax-Wendroff is somewhat better than Beam-Warming and that Lax-
Wendroff produces oscillations behind the shock while the oscillations in the Beam-Warming
are located in front of the discontinuity.
Below, we will see how to design high resolution or non-oscillatory scheme which in principle
are second order accurate in regions where the solution is smooth and only first ordernon
oscillatory near the shocks.
Third, note that the leap-frog scheme seem to the exhibit the worst oscillatory behaviour for
the non-smooth solution. In fact, most of this is due to the presence of the computational
mode which tend to amplify the oscillations. Things look a lot better when the the compu-
tational mode is controlled by the Robert-Asselin filter. Notice that the filtered leap-frog is
somewhat between the Lax-Wendroff and a first order method for it exhibits some oscillations
behind the shock and smooths out the discontinuity at the same time.

The matlab programs used to generate those results can be obtained upon request from the author.

Below, we analyze the different schemes in detail to understand better their performances.

4.8 Numerical diffusion, dispersion, and the modified equation

Consider the upwind scheme (40) for the advection equation. With some simple manipulations this
scheme can be rewritten as
n+1
1 t uj 2unj + ujn1 unj+1 unj1 ax unj+1 2unj + unj1
(un+1
j ujn1 ) + = a +
2t 2 (t)2 2x 2 (x)2
or
un+1
j ujn1 unj+1 unj1 n n n n+1 n n1
ax uj+1 2uj + uj1 t uj 2uj + uj
+a = + . (51)
2t 2x 2 (x)2 2 (t)2
For all practical purposes this scheme can be thought of as an approximation for the diffusive
advection equation
ax t
ut + aux = uxx + utt ,
2 2

38
with a second order truncation error, O((x)2 + (x)2 ). Differentiating the advection equation
once with respect to t, yields utt = auxt = (aut )x = a2 uxx , i.e, the diffusive equation above becomes
 
ax at
ut + aux = 1 uxx .
2 x
This equation is sometimes called the modified equation approximated by the upwind scheme to
the second order. In essence, the upwind scheme approximates the viscosity solution introduced in
(34) corresponding to the advection equation with the viscosity = ax
2 (1 ) where = at/x
is the Courant number. On the one hand this can make us believe that in fact the upwind scheme
computes the right physical solution, by approximating the vanishing viscosity solution, and on the
other hand it provides an explanation on the poor performance of this scheme, especially in the
case of the smooth solution in Figure 10, of under-predicting the solution.

Note the viscosity coefficient, introduced by the upwind scheme, is zero if the Courant number if
choosing to be perfectly one. Such choice is avoided in practice because it may lead to numerical
instabilities because of round-off errors. Further as noted above a little viscosity is necessary to
guarantee convergence to the physical solution.
n n n
ax uj+1 2uj +uj1
This phenomenon is known as numerical dissipation or diffusion. It is due to the term 2 (x) 2 +
n+1 n1
t uj 2unj +uj
2 (t)2
on the right hand side of (51). In fact, multiplying the viscous advection equation
by u and integrate on [0, 1] yields

d 1 u2 a 1 2
Z Z Z 1
dx + (u )x dx = uuxx dx. (52)
dt 0 2 2 0 0

Where we used uut = (u2 )t /2 and uux = (u2 )x /2 . By integration by parts we have
Z 1 Z 1 Z 1
1 2
uuxx dx = uux |0 (ux ) dx = (ux )2 dx
0 0 0

provided the boundary terms vanish, which happens if we use periodic or homogeneous Dirichlet or
Neumann boundary conditions. The second term on the right of the integral equation (52) vanishes
for the same reason and yields the following energy dissipation principle

d 1 u2
Z Z 1
dx = (ux )2 dx < 0 if ux 6= 0.
dt 0 2 0

Thus if integrated to infinity the upwind scheme will smooth out all the gradients in the solution
and will ultimately converge to a constant-flat solution.

As we can surmise from Figures 10 and 11, the Lax-Friedrichs scheme is more dissipative than the
upwind scheme. Indeed, the Lax-Friedrichs scheme is equivalent to

un+1
j unj unj+1 unj1 n n n n+1 n n1
(x)2 uj+1 2uj + uj1 t uj 2uj + uj
+a = . (53)
t 2x 2t (x)2 2 (t)2
which yields a dissipative modified equation, which is the second order approximation of the Lax-
2
Friedrichs scheme, with a viscosity = (x) 2 t 2 2
2t a 2 = (x) (1 )/(2t) which is typically
larger than that of the upwind scheme.

39
This dissipation or diffusivity problem is typical to first order schemes for hyperbolic systems. Let
us now consider the second order schemes and derive the modified equations associated with those
schemes. Since they are already second order accurate, we should consider approximations with an
order of accuracy higher than two. Below, we take a slightly different route than it is done above,
to derive the modified equation. Namely, we use Taylor expansion as to compute the truncation
error but instead we keep the first neglected term and add it to the advection equation to form the
modified equation.

For the leap-frog method we have

un+1
j ujn1 unj+1 unj1 (t)2 (x)2
+a = ut + uttt (x, t)+O((t)4 )+aux +a uxxx (x, t)+O((x)4 )
2x 2x 6 6
(because all even terms in the Taylor series cancel out). Again using the advection equation we
have: uttt = a3 uxxx and therefore the modified equation, which is approximated to the fourth
order by the leap-frog scheme, is

a(x)2
1 2 uxxx .


ut + aux = (54)
6
This equation is known as the dispersive wave equation. First, it is easy to show that this equation
conserves energy (see exercise 11 below). This is consistent with the fact that the von Neumann
amplification factors for the leap-frog method satisfy | | = 1 when < 1.

Second, dispersion refers to a physical system where waves of different wavelengths propagate at
different wave speeds. Lets look at wave like solution on the form

u = ei(kxt)

for the dispersive wave equation. Here k = 2l is the wavenumber, is called the phase or
frequency, /k defines the phase speed, the speed at which the wave propagates. Plugging this
ansatz into (54) yields, the dispersion relation

a(x)2
1 2 k 3 ,


= ak
6
2
i.e the phase speed, ck = a a(x) 1 2 k2 , depends highly on the wavenumber, and higher the


6
wavenumber faster the wave disturbance moves relative to the advective motion. Short waves move
relatively faster. This is very unlike the plain-original advection equation were all waves move at
the same speedthe advection speed a.

Now, we reconsider the von Neumann amplification factors for the leap-frog method in (45). Recall
that we have two solutions (one is physical and the other is a numerical artifact), which we denote
here by:
u+ = ei(kjxn0) ; u = (1)n ei(kjx+n0 ).
Let us consider only the physical mode, u+ . From the discussion above, using n = tn /t, the
associated phase speed is given by

0 arcsin( sin(l ))
c+ = =
kt kt

40
Using Taylor expansion to expand both the sine and the arcsine functions, for small l values
(resolved modes), yields

(x)2
0 l (1 2 )3l = atk at(1 2 )k3
6 6
which implies that the phase speed of the physical mode for the leapfrog scheme satisfies

(x)2
c+ a a (1 2 )k2 .
6
This matches exactly the expression for ck found above, using the dispersive wave equation.

Now, we can explain why in Figure 11, the leapfrog scheme exhibits oscillations. They result from
the fact that the small truncation errors located near the discontinuity, which also accumulate with
time, are viewed as wave disturbances of much shorter wavelengths which then propagate at their
own speeds different from that of the actual solution.

Recall that the amplification factors for the previous two first order schemes, namely the upwind
and the Lax-Friedrichs, are respectively given by

uw = 1 (1 cos(l )) i sin(l ) and LF = cos(l ) i sin(l ).

In the light of the analysis done in the previous paragraphs for the leap-frog scheme, we set n
eint where is a generalized phase so that ()/k yields the phase speed of the numerical wave
solution and () yields the exponential growth or damping rate of the wave amplitude. We have
for the upwind and Lax-Friedrichs schemes, respectively,

eiuw t = 1 (1 cos(l )) i sin(l ) , eiLF t = cos(l ) i sin(l ).

For small t, we have

eit 1 it.
Hence
uw t sin(l ) i(1 cos(l )) and LF t sin(l ) i(1 cos(l )).
First note that both schemes have quite strong damping rates,

1 1 cos(l ) a
(uw ) = (1 cos(kx)) k2 x, (LF ) = xk2 ,
t 2 t 2
respectively, and consistently with the previous results, the LF scheme being more dampedby a
factor of 1/.

With l = kx, the phase speeds are equal and are given by

k2 (x)2
 
(up ) (LF ) a
= = sin(kx) a 1 .
k k kx 6

This clearly shows that both schemes are dispersive, similarly to the leap-frog scheme. Nevertheless,
small wave disturbances typically occurring at the grid scale decrease in magnitude at a faster rate
than the domain-scale physical wave.

41
Now we turn to the Lax-Wendroff and Beam-Warming schemes. We have

LW = 1 i sin(l ) 2 (1 cos(l ))

yielding
2
LW = sin(l ) i (1 cos(l )), (55)
t t
i.e, for the phase speed, we have the same dispersive behaviour as for the previous schemes but a
much smaller damping rate

2 2 2
(wLW ) = (1 cos(l )) k (x)2 ,
t 2t
of the same order as the phase speed deviation. For grid scale disturbances, the wavenumber, k,
2
scales as 1/x, therefore the damping rates for those small scale waves is on the order O( 2t )
which seems to be small, explaining the propagation of the small oscillations, in Figure 10, away
from the discontinuity before they get damped.

More importantly note the propagation of the oscillations for the Lax-Wendroff scheme, in Figure
10, to the left of the discontinuities, is associated with the fact that according to the dispersion
relation (55) smaller wavelength disturbances move slower than those with larger wavelengths. We
have a similar behaviour for the Lax-Wendroff scheme. But the Beam-Warming scheme exhibits
somehow larger amplitude oscillations moving to the right, which seems to anticipate that the
Beam-Warming scheme has both a weaker damping rate and dispersive phase speeds characterized
by smaller wavelengths propagating faster than larger ones. The details are left as an exercise.

Exercise 11 Show that the dispersive wave equation

ut + aux = uxxx

conserves energy, that is

1
d u2
Z
dx = 0.
dt 0 2

Exercise 12 Determine the exponential damping rate and the phase speed for the Beam-Warming
scheme. Deduce that this scheme is dispersive and that wave disturbances with larger wavenumbers
(smaller wavelengths) propagate faster than those with smaller wavenumbers.

42
 1
Initial
Exact
0.9 upwind
LaxFriedrichs

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0.2 0 0.2 0.4 0.6 0.8 1 1.2

Figure 12: Riemann Problem for Burgers equation solved with both upwind and Lax-Friedrichs
schemes, showing that the upwind schemes predicts a wrong shock speed.

5 Finite volume methods for scalar conservation laws

5.1 Wrong shock speed and importance of conservative form

Consider Burgers equation with the discontinuous initial condition

1
ut + (u2 )x = 0
 2
1 if x < 0.25
u0 (x) = (56)
0 if x > 0.25.

We view Burgers equation as an advection equation with a non-linear advection speed a(u) = u
and attempt to solve this problem with both the upwind and Lax-Friedrichs methods. The upwind
scheme is generalized to Burgers equation as follows.
t
un+1 = unj max(unj , 0)(unj unj1 ) + min(unj , 0)(unj+1 unj ) .


j (57)
x
Whereas the formula for Lax-Friedrichs scheme amounts to approximating the flux derivative using
centred differences.
1 t
un+1 = (unj+1 + unj1 ) (unj+1 )2 /2 (unj1 )2 /2 .


j (58)
2 2x

The results are shown in Figure 12 where the two numerical solutions are compared to the exact
solution to the Riemann problem. Note that the Lax-Friedrichs method predicts well the propaga-
tion of the shock wave, with a significant smearing of the discontinuity, as excepted, due its high
viscosity, while the upwind scheme predicts a steady solution which doesnt change with time. In
fact, this latter result can be easily recovered analytically from the upwind scheme. The numerical
solution in this case remains zero on the right of the shock because the advection velocity is zero
and remains one on the left because the backward finite difference derivative is zero.

43
The main reason for why the Lax-Friedrichs scheme out performs the upwind scheme, in this case,
is because the former has a conservative form. A numerical scheme for a conservation law
ut + (f (u))x = 0
is said conservative if it can be written on the form
un+1 = unj + Fj+1/2 Fj1/2 .
 
j (59)
Where Fj+1/2 , called a numerical flux which is in some sense an approximation of the flux f (u) at
the cell interface, xj+1/2 = xj + x/2.

Here we show that indeed the Lax-Friedrichs scheme has a conservative form. Let
LF 1
Fj+1/2 = (unj+1 unj ) (f (unj+1 ) + f (unj ).
2 2
Then the Lax-Friedrichs schemes can be written as
un+1
j = unj + Fj+1/2
LF LF
Fj1/2 .
Thus it is conservative. We will see below that, the Lax-Wendroff and the Beam-Warming methods
are also conservative.

Remark:
It is worthwhile noting that the formula for the Lax-Friedrichs flux, F LF , at the interface j + 1/2,
is the average of the left and right fluxes, f (uj ), f (uj+1 ), plus a centered difference, uj+1 uj ,
suggesting a diffusive term. Thus, the Lax-Friedrichs scheme can be viewed as a discrete version of
the diffusive equation
ut + (f (u) ux )x = 0
(x)2
where = 2t is the viscosity coefficient.

Convergence of Conservative Schemes:

It is shown in the literature that if a numerical scheme for a conservation law ut +(f (u))x = 0
is consistent, stable, and has a conservative form, then the resulting numerical solution
converges to a weak solution to the conservation law.

This statement can be viewed as an extension of the Lax-equivalence theorem to non-linear con-
servation laws. Notice however, since weak solutions are not unique, it is not guaranteed that
the numerical solution obtained by a consistent, stable, and conservative scheme converges to the
physical entropic solution. For that some extra-properties of the numerical scheme are needed to
enforce the entropy condition.

5.2 Godonuvs first order scheme

Let xj = jx, j = , 2, 1, 0, 1, 2, and tn = nt, n = 0, 1, 2, be a discretization of the

space-time domain (x, t). We divide the real line into sub-intervals [xj1/2 , xj+1/2 ], called grid cell,
where xj+1/2 = xj + x/2. We define the cell average of the solution u(x, t) at each grid cell as
Z xj+1/2
n 1
uj = u(x, tn )dx. (60)
x xj1/2

44
Consider the conservation law
ut + (f (u))x = 0.
Next, we integrate this equation on the rectangle [tn + tn+1 ] [xj1/2 , xj+1/2 ], often called the
control or finite volume, hence the name finite volume method. We have
Z tn+1 Z xj+1/2 Z tn+1 Z xj+1/2
ut (x, t)dxdt + (f (u(x, t)))x dxdt = 0
tn xj1/2 tn xj1/2

or
Z xj+1/2 Z xj+1/2 Z tn+1 Z tn+1
u(x, tn+1 )dx u(x, tn )dx+ (f (u(xj+1/2 , t)))dt (f (u(xj1/2 , t)))dt = 0.
xj1/2 xj1/2 tn tn

Dividing by the mesh size x and introducing the averages unj above, yields

t h n i
un+1
j = unj n
Fj+1/2 Fj1 , (61)
x
where
tn+1
1
Z
n
Fj+1/2 = (f (u(xj+1/2 , t)))dt,
t tn
is the average flux of u through the interface x = xj+1/2 , tn t tn+1 . Provided we find an
adequate numerical approximation to the time integral, the formula (61) provides a numerical
scheme for the conservation law, in conservative form.

Piecewise constant approximations and the Riemann problem at the cell interfaces:
Assume that at time t = tn , the cell averages unj are known. Consider the approximation:
u(x, tn ) unj , if xj1/2 x xj+1/2 .
Then finding an approximation for the flux Fj+1/2 , amounts to solving the Riemann problem

ut + (f (u))x = 0, t [tn , tn+1 ] (62)


uL if x < xj+1/2
u(x, tn ) =
uR if x > xj+1/2
where the left and right states are given, respectively, by uL = uj and uR = unj+1 .

To compute the flux Fj+1/2 we need to know the solution u along the interface x = xj+1/2 , tn
t tn + t. We introduce the shock speed at the cell interface,
f (uR ) f (uL )
s = ,
uR uL
according to the Rankine-Hugoniot jump condition. Under the condition that f (u) is convex, using
the method of characteristics, introduced above, we have, for tn t tn + t,
uL if f (uL ) > 0 & f (uR ) > 0


uR if f (uL ) < 0 & f (uR ) < 0



u(xj+1/2 , t) = uL if f (uL ) 0 & f (uR ) 0 & s > 0 (63)

uR if f (uL ) 0 & f (uR ) 0 & s < 0



u if f (uL ) 0 & f (uR ) 0,

45
where in (63), u , called a sonic point, is defined such that f (u ) = 0 and corresponds to a
rarefaction wave solution. Notice that the convexity of f guarantees that u exists and is unique.
Note also the fist four cases correspond either to an entropic shock solution where, according to
Laxs criterion,
f (uL ) s f (uR )
so that uj+1/2 = uL if s > 0 and uj+1/2 = uR if s < 0 or a rarefaction waves where the rarefaction
fan is either completely to the left or completely to right of the interface according to whether
f (uR,L ) < 0 or f (uR,L ) > 0, respectively. The last case, however, corresponds to rarefaction wave
where the rarefaction fan contains characteristics going both to the left and to the right. It is called
a transonic rarefaction, by analogy to gas dynamics, where such a rarefaction happens when the
fluid on one side of the wave moves at a speed smaller than the speed of sound (subsonic) and on
the other side it moves with a speed larger than the speed of sound (supersonic).

Here we show that the sonic point u is in fact the root of the equation f (u ) = 0. Assume
f (uL ) < 0 < f (uR ), then in this case there is a rarefaction wave connecting the left and right
states. Inspired by the solution to Burgers equation, a rarefaction wave solution is sought on the
form u(x, t) = v(x/t). Plugging this ansatz into the conservation law ut + (f (u))x = 0, yields
x 1
v (x/t) + f (v(x/t))v (x/t) = 0.
t2 t
Hence
x
f (v(x/t)) =
.
t
Along the interface x = 0 we have u = u(0, t) = v(0) and f (v(0)) = 0.

The celebrated Godunovs method is now obtained by simply using the solution to the Riemann
problem in (63) at each interface xj+1/2 to computes the numerical fluxes Fj1/2 , Fj+1/2 in (61),
yielding

if f (uL ) > 0 & f (uR ) > 0

f (uL )
if f (uL ) < 0 & f (uR ) < 0

(uR )
f


Fj+1/2 = f (uL ) if f (uL ) 0 & f (uR ) 0 & s > 0 (64)
f (uR ) if f (uL ) 0 & f (uR ) 0 & s < 0




f (u ) if f (uL ) 0 & f (uR ) 0.

Notice that the solution u in (62) can be replaced by the shock solution, i.e, uR if s < 0 and
uL if s > 0, even when f (uL ) < 0 < f (uR ), and will still provide a weak solution satisfying the
Rankine-Hugoniot jump condition, but as we already know this will lead to an unphysical weak
solution which violates the entropy condition. The resulting numerical solution is referred to as
an all shock solution and provides an approximation to the weak solution to the conservation law,
which is not necessarily the physically relevantentropic solution.

Case of the advection equation and Stability of Godunovs method

If we replace the conservation law by the simple advection equation with a positive advection speed,
a > 0, then the solution to the Riemann problem at the cell interface, xj+1/2 , reduces to
u(xj+1/2 , t) = uj ,

46
 1
Initial
Exact
0.9 Godunov
LaxFriedrichs

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0.2 0 0.2 0.4 0.6 0.8 1 1.2

Figure 13: Same as Figure 12 but with Godunovs method instead of the upwind scheme.

and Godunovs method reduces to the upwind scheme

at n
un+1 = unj (u unj1 ). (65)
j
x j
Recall that the upwind method is stable under the CFL condition. Therefore, provided the CFL
condition
max(f (uL ), f (uR ))t x (66)
is satisfied, Godunovs method is stable.

Consistency and convergence

Moreover, given the way the fluxes were computed, i.e, using the exact solution of the Riemann
problem using first order approximation of the initial data, namely piecewise constant cell averages,
we can show that Godunovs method is consistent of order 1 in both space and time, just like the
upwind scheme. Moreover, since it is also conservative by construction, Godunovs method is
guaranteed to convergence to the entropic weak solution, provided we always choose the entropic
solution for the Riemann problem. The numerical test of Figure 12, is repeated with the Godunov
method and the results are shown in Figure 13, where Godunovs method is compared to the
Lax-Friedrichs method. As we expect Godunovs method predicts the right shock speed and has
much less dissipation than the Lax-Friedrichs method. However, one big disadvantage of Godunovs
method is that it requires the solution of a Riemann problem at each cell interface and every time
step. This can be very costly especially for non-linear systems of conservation laws.

An other serious shortcoming of Godunovs method is that it is only first order in both time
and space. This limits its use in practice but it constitutes an important cornerstone for more
sophisticated so called high resolution methods developed below.

5.3 High resolution, TVD, and MUSCL schemes

As noted above one of the major shortcomings of Godunovs method is that it is only first accurate
in both space and time. As noted above a crucial step in deriving this method is the piecewise

47
constant approximation step, when the solution u(x, t) at time t = tn is approximated by the cell
average
u(x, tn ) unj , xj1/2 x xj+1/2 ,
to advance to the next time step, tn+1 , which is only a first order approximate/interpolation. Since
Godunovs method computes the cell averages at each new time step, this piecewise approximation
appears to be very convenient, indeed. Given the cell averages unj , we can use piecewise linear
interpolation to yield a second order reconstruction
u(x, tn ) u(x, tn ) unj + j (x xj ), xj1/2 < x < xj+1/2 ,
which can be used instead to advance the solution to next step (see Figure 14). The slopes j can
be computed in many different ways; using many different finite differencing formulas as we will
see below.

To illustrate let us consider the advection equation

ut + aux = 0, a > 0.
Given the piecewise linear reconstruction values, u(x, tn ), at time t = tn , the exact solution for
this equation at t = tn + t is given by
u(x, tn + t) = u(x at, tn ).
Provided at/x < 1, we have
 n
uj1 + j1 (x at xj1 ) if xj1/2 < x < xj1/2 + at
u(x, tn + t) = (67)
unj + j (x at xj ) if xj1/2 + at < x < xj+1/2 .

tn+1 ) on the intervals (xj1/2 , xj+1/2 ),

Averaging u(x,

at n j1 xj1/2 +at
Z
un+1 = u + (x at xj1 )dx+
j
x j1 x xj1/2
Z xj+1/2
x at n j
uj + (x at xj )dx,
x x xj1/2 +at

yields the cell averages at tn + t, given by

at n at
un+1 = unj (u unj1 ) (x at) (j j1 ). (68)
j
x j 2x

If we set the slopes to zeros, j = 0, then we recover the first order Godunov (upwind) method in
(65). A second order scheme is obtained, if the slope is choosing to be a first order finite differencing
formula which approximates the derivatives ux (xj , tn ) in the jth grid cell, using neighbouring cell
average values. The following three choices of slopes yield three popular second order schemes
uj+1 uj1
Centered: j = (Fromm)
2x
uj uj1
Upwind: j = (Beam-Warming)
x
uj+1 uj
Downwind: j = (Lax-Wendroff).
x

48
Exercise 13 Show that when a < 0 the second order scheme for the advection equation using
piecewise linear reconstruction is given by
at n at
un+1 = unj (u unj ) + (x + at) (j+1 j ). (69)
j
x j+1 2x

Reconstruct, Solve, Average:

The algorithm followed above to derive Godunovs method as well as its second order version for
the advection equation involves three main steps.

1) Reconstruct: Given the cell averages unj , we (re)construct piecewise constant or piecewise
linear approximations

u(x, tn ) ujn or u(x, tn ) ujn + j (x xj ), xj1 < x < xj+1/2

2) Solve: Solve the Riemann problem

ut + (f (u))x = 0, tn t tn+1 , u(x, tn ) = u(x, tn )

3) Average: Average the solution at tn+1

Z xj+1/2
n+1 1
uj = u(x, tn+1 ) dx.
x xj1/2

A few remarks on the Reconstruct-Solve-Average algorithm:

Note the first step is crucial, it sets the order of accuracy of the method. Piecewise constants yield
a first order method while a piecewise linear approximation yields a second order scheme. In fact
methods using parabolic approximations exist are often used in practice, they are known are PPM
(for piecewise parabolic methods). Piecewise parabolic reconstruction yields a third order method
and so on. For, the second step it is desirable to be able to solve the conservation law exactly.
This is possible in the case of the advection equation and for the conservation law in general when
piecewise constants are used in step one, yielding Godunovs method. But in general, we need to
resort to quadrature formulas to approximate the flux integrals along the time interval [tn , tn + t].

Finally, when a finite volume method, such as Godunovs, is applied to a conservation law, it yields
the average values un+1
j directly, which takes care of step three.

High order finite volume method for conservation laws

Recall the finite volume derived in (61) for Godunovs method

t h n i
un+1
j = unj n
Fj+1/2 Fj1 ,
x
where
tn+1
1
Z
n
Fj+1/2 = (f (u(xj+1/2 , t)))dt.
t tn

49
To reconcile with the case when u(x, t) is not piece-wise constant at time t = tn , we approximate
the time integral by a rectangular rule yielding an Euler step
t  n 
uj+1 = unj n
Fj+1/2 Fj1/2 (70)
x
where
n
Fj+1/2 = f (u (xj+1/2 , t+
n )).
Here
u (xj+1/2 , t+
n ) = lim u(xj+1/2 , t)
ttn
is obtained by solving the Riemann problem with left and right states
uL = u(xj+1/2, , tn ), uR = u(xj+1/2,+ , tn )
where again
u(xj+1/2, , tn ) = lim u(x, tn ) and u(xj+1/2,+ , tn ) = lim u(x, tn ).
xxj+1/2,x<x xxj+1/2,x>x
j+1/2 j+1/2

Note that in the case of a piecewise constant approximation these limits are simply uL = unj and
uR = un+1
j , and the solution to the Riemann problem is constant on the time interval [tn , tn + t],
provided the CFL condition t max(|f (uR )|, |f (uL )|) x is satisfied. Note this is not true for
higher order reconstructions.

To obtain higher order accuracy in time, we often use a predictor-corrector scheme such as the
mid-point Runge-Kutta method in place of the first-order Euler step (70):

t  n 
uj+1/2 = unj n
Fj+1/2 Fj1/2 (71)
2x
n t  n+1/2 n+1/2

uj+1 = uj Fj+1/2 Fj1/2
x
where
n+1/2
Fj+1/2 = f (u (xj+1/2 , t+
n+1/2 )).

Oscillations and TVD methods

Recall from Figure 11 that one major problem with second order methods is that they tend to
generate unphysical oscillations near discontinuities. One way to control these oscillation is to use
a hybrid method where the second order reconstruction is limited to regions where the solution, u,
is smooth and use a first order method in regions where u presents discontinuities of some sort. To
find an intelligent way to do so, we first introduce a mathematical tool to control those oscillations.
This is achieved by the total variation, which is introduced next.

Definition 6 The total variation (TV) of a real valued function f is given by

+
X
TV(f ) = sup |f (j+1 ) f (j )|
j=

where the supremum is taken on all subdivisions < 1 < 0 < 1 < of the real line.

50
To help develop some intuition, we list a few properties of TV.

If f (x) is monotonic (i.e, non-increasing or non-decreasing) on some interval [a, b], then the
total variation of f on [a, b] is given by

TV[a,b] (f ) = |f (b) f (a)|

If f (x) is piecewise constant on the line segments (xj1/2 , xj+1/2 ), then

+
X
TV(f ) = |f (xj+1 ) f (xj )|,
j=

i.e, the sum of all the jumps of f .

If f (x) is piecewise linear on the line segments (xj1/2 , xj+1/2 ), then

+
X +
X
+
TV(f ) = |f (x
j+1/2 ) f (xj1/2 )| + |f (x+
j+1/2 ) f (xj+1/2 )|,
j= j=

i.e, the sum of the variations within each one of the line segments plus all the jumps across
the interfaces.

If f is differentiable, then Z +
TV(f ) = |f (x)|dx.

Since the exact solution to the advection equation simply propagates at a constant speed, its TV
doesnt change with time. Moreover, it can be shown that the total variation for an entropic
solution for a conservation law, in general, doesnt increase with time. It can decrease after a shock
but never increases. However, this is not always the case for the numerical solution. Clearly the
oscillations generated by the second order schemes in Figure 11 do increase the TV of the numerical
solution. A reasonable way to attempt to avoid those unphysical oscillations, is thus to require that
the total variation of the numerical solution does not increase.

Definition 7 The numerical scheme

t  n 
uj+1 = unj n
Fj+1/2 Fj1/2
x
is said to be total variation diminishing (TVD) if

TV(un+1 ) TV(un ). (72)

For instance we can show that the average step in the Reconstruct-Solve-Average algorithm does
actually diminish total variation. Therefore, when the slopes are set to zeros, j = 0, the upwind
scheme for the advection equation is TVD.

51
Definition 8 A numerical scheme
un+1
j = H(unjk , unjk+1 , , unj+l )
is said to be monotone if
H
0, = j k, , j + l.
u

One important property of monotone schemes is that they do not create new local maxima or
minima:
max un+1
j max unj , and min un+1
j min unj
As such monotonic schemes are TVD. On the other hand TVD schemes are monotonicity preserving.
A numerical scheme is said to be monotonicity preserving if monotonic data remains monotonic:
unj1 unj unj+1 = un+1 n+1
j1 uj un+1
j+1

We have the following general statement

monotonic schemes TVD schemes monotonicity preserving schemes.

It is easy to show that, under the CFL condition, at x, Godunovs method for the advection
equation is monotonic. In fact, for a > 0, we have
un+1
j = H(unj , unj1 ) (1 )unj + unj1 .
Therefore it is TVD as noted above. Unfortunately, it is not possible to construct linear second order
schemes which are monotonic, according to the following theorem due to Godunov. In agreement
with the fact that second order schemes exhibit oscillations.

Theorem 6 (Godunov) A linear monotonic scheme is at most first order accurate.

In order to achieve high order schemes without unphysical oscillatory behaviour we need to make
schemes which are intrinsically nonlinear. A family of such schemes are the hybrid schemes men-
tioned above, which are second order in smooth regions and only first order near shocks, to avoid
oscillations.

To guarantee that our hybrid numerical scheme remains TVD and second order accurate in smooth
regions, we must choose the slopes j so that the TV of the reconstructed function is not larger
than that of the discrete cell averages unj . This is achieved by defining the slope j as a non-linear
function of the data unj in such a way to control the total variation. The associated non-linear
function is called a limiter and methods based on this idea are known as slope-limiter methods,
first introduced by van Leer when he derived his famous MUSCL schemes (Monotonic Upstream-
Centered Schemes).

Perhaps the simplest choice of a limiter, which guarantees second order accuracy in regions where
u is smooth and satisfies the TVD property at the same time, is the minmod limiter:
unj+1 unj unj unj1
j = minmod( , ) (73)
x x

52
where
a if |a| < |b| and ab > 0
minmod(a, b) = b if |b| < |a| and ab > 0
0 if ab 0.

Note that when a, b have the same sign, the minmod function chooses the one which is smaller in
magnitude and when ab 0, it returns zero. Instead of using the downwind slope, yielding the
Lax-Wendroff scheme, or the upwind slope, leading to the Beam-Warming scheme, the minmod
limiter chooses the one which is smaller in magnitude. When the upwind and downwind slopes
have opposite sign, the reconstructed cell value is kept constant. This latter must correspond to a
local minimum or a local maximum, and the minmod limiter tends to preserve the local extrema,
hence avoiding to create overshootings and undershootings near those extrema. Thus it does not
increase TV and does not generate oscillations.

A more popular choice of limiter, due to van Leer, is the MC-limiter (monotonized-centered
limiter):
unj+1 unj1 unj+1 unj unj unj1
j = minmod( ,2 ,2 )
2x x x
Note, the MC-limiter chooses the smallest in magnitude between the centred-differences, corre-
sponding to Fromms method, and twice the upwind or the downwind formulas, and returns zeros
when there is a change is sign. This limiter yields highly accurate centered slopes in smooth regions
and sharper resolution near discontinuities.

Flux-limiters
Using piecewise linear reconstruction for the advection equation yields (see (68) and (69))
t n
un+1
j = unj (F n
Fj1/2 )
x j+1/2
where a
aunj + (x at)j if a > 0
2


n
Fj+1/2 =
aunj+1 a (x + at)j+1

if a < 0.

2

Introducing the negative and positive parts of a

a+ = max(a, 0), a = min(a, 0)

the flux Fj+1/2 can be rewritten in a compact form as

 
+ 1 t|a|
Fj+1/2 = a uj+1 + a uj + |a| 1 j , (74)
2 x
where j is the jump in u across the upwind interface; j = uj uj1 if a > 0 and j = uj+1 uj
if a < 0.

We introduce the upwind side of j as


j1 if a > 0
J=
j+1 if a < 0

53
 Piecewise linear
Piecewise constant

j1 j j+1

Figure 14: Piecewise constant (dashed) and piecewise linear reconstructions.

and the jump

uj = uj+1 uj .
We define

uJ
= .
uj
and let
j = ()uj

in (74), where is a non-linear function of , called a flux-limiter, which is defined so that the
resulting scheme is second order accurate in smooth regions and TVD at the same time.

Some linear choices of yield the popular second order (linear) schemes:

() = 0 : upwind
() = 1 : Lax-Wendroff
() = : Beam-Warming
1
() = (1 + ) : Fromm,
2
which are of course not TVD according to Godunovs theorem. While some clever non-linear choices
yield high resolutionTVD methods. The following are the most popular ones:

() = max(0, min(1, )) : minmod

() = max(0, min((1 + )/2, 2, 2)) : MC-limiter
() = max(0, min(1, 2), min(2, )) : superbee
1+
() = : van Leer.
1 + ||

54
They all have their strengths and weaknesses, depending on the application. Those are well docu-
mented in the literature (see LeVeques book: finite volume methods for hyperbolic problems, for
examples). The performance of each of the choices of limiters listed above is shown on the different
panels in Figure 15, for both the smooth and the non-smooth/square wave. We see that in general
the TVD limiters capture very well both the smooth solution (with an accuracy comparable to
the second order schemes) and the non-smooth square wave. The second order methods have an
oscillatory behaviour near the discontinuity and the upwind method is too diffusive.

References

[1] Crispin W. Gardiner, 2004, Handbook of Stochastic Methods: for Physics, Chemistry and the
Natural Sciences, Springer.

[2] Randall J. LeVeque, 2002, Finite Volume Methods for Hyperbolic Problems (Cambridge Texts
in Applied Mathematics),Cambridge University Press.

[3] Dale Durran, 1998, Numerical Methods for Wave Equations in Geophysical Fluid Dynamics,
Springer.

55
 2 2
Exact Exact
minmod MC
1.5 1.5

1 1

0.5 0.5

0 0
0.5 0 0.5 1 0.5 0 0.5 1

2 3
Exact Exact
1.5 superpee 2 van Leer

1 1

0.5 0

0 1
0.5 0 0.5 1 0.5 0 0.5 1

2 3
Exact Exact
1.5 upwind 2 LaxWendroff

1 1

0.5 0

0 1
0.5 0 0.5 1 0.5 0 0.5 1

3 3
Exact
Exact
Fromm
2 BeamWarming 2

1 1

0 0

1 1
0.5 0 0.5 1 0.5 0 0.5 1

Figure 15: Performance of the different limiter functions listed in the text for both the smooth and
non-smooth data.

56